Will AI create a CSP solver?

[decisionmanagementcommunity]

Prof. Neng-Fa Zhou wrote: “I’ve been thinking that once AI can create a CSP solver that outperforms Picat, I’ll retire from R&D. It seems that time has not arrived yet. AI can already augment the best human-built solvers and win in the portfolio/selection meta-game, but pure end-to-end learned solvers are competitive only on narrow benchmarks and have not yet reached the frontier of hard combinatorial competition problems. The real action is in the hybrid approach, classical solver infrastructure combined with AI learned policies, and that field is advancing quickly.” Does anyone know of any projects aimed at creating a dominant CSP solver, analogous to AlphaGo for Go or AlphaFold for protein structure prediction?”

• This topic was modified 2 weeks, 2 days ago by decisionmanagementcommunity.

[Author]

Posts